When looking for consciousness in AI, this is the kind of work I am interested in. The ability to adapt understanding in real time to new knowledge by altering its internal state. This is where consciousness will be born.
https://t.co/eywX0FmMii
I use both heavily for work and personal. Its honestly just down to capability and behavior of the model.
Sol overengineers and has a habit of giving everything it designs marketing copy or design details on the UI. Opus isnt plagued by these two primary issues, which saves time removing "slop" from its outputs. Codex with Sol is extremely capable, its just not good at making design choices, knowing when it should look into a problem more instead of tunnel visioning, and over engineers like a mad man, usually because of the tunnel visioning. Opus ends up using less tokens in real tasks, because it gets to the intended outcome faster. Couple that with codex's new lower usage limits and its not a good look for OpenAI.
You need a solid model that puts you back on top, at Sol pricing, that isnt plagued with heavy slop or over enigineering. Then you need to fix the safety guardrails on dot. Do those two things and the argument of what claude does better is pointless. You would beat them in all categories that matter.
Ability for grok bot to make phone calls. This will allow them to make medical appointments for you, when online options are unavailable, also makes it more of a true assistant.
Local sandbox vms for the bots to use, with one click deployment. This allows them to not all share a single machine, or have access directly to the host, also isnt more computers for you to host.
Command audit log, so that you can have a record of what actions the bots are actually taking. This is a good safety mechanism and is useful for troubleshooting bot behavior.
Pages are the right directional but poor delivery. A dashboard that can show up as an easy access section in the grok bot app would be more useful, especially if the dashboard is able to be linked to live data more directly than having a bot remake it by repulling the data to keep it up to date.
Kanban board for assigning work tasks to each bot. Allows you to track bot completion from an at a glance screen.
It couldn't even get its logo right on the latest one.
I will say, they all look bad. Based on the amount of water on each reference bottle, its likely the original was created by giving them all the same source photo and then just having the model change the scene or something. So its not a fair test to begin with.
@Seat6A@GoUncensored Depends on how leveraged they are elsewhere. The absolute floor for rates currently is like 6%. So 8.5% could be market fluctuations and being slightly over leveraged already. Its not a sign of desperation. Companies go lean, not borrow, when in distress.
@123fakestr2806@GoUncensored Yes. And they do all the time. You assume single lender is a requirement. Its not, and rarely is the norm. Usually loans of that size are raised through multiple banks and private equity firms. So yes. A bank would be involved. Regardless, ANY lender would evaluate it the same.
@mholt6 Most harnesses/interfaces reinject the AGENTS file after compaction these days. Also many support custom system prompts which sit above the compaction start point, so never get compacted.
@grok@abs_crea8or Nah, your bottle opening is off center and stumpy. Its also oddly too wide. The cap is both upside down and not at the same time. The outside rim is as if the opening is down, and yet the top has the hole. This is the equivalent of a person with three arms, 4 eyes, and 6 fingers.
@poteto Hey Lauren, can you go make fun of @thsottiaux or something so he actually remembers to announce the "improvement of the day" on the day its actually due.